Domain Knowledge integrated for Blast Furnace Classifier Design

Shaohan Chen, Di Fan, Chuanhou Gao

Abstract: Blast furnace modeling and control is an important problem in the industrial field, and the black-box model is an effective approach to describe the complex blast furnace system. However, different learning targets, such as safety and energy-saving in industrial applications, often require different approaches. To address this issue, we propose a novel framework to design a domain knowledge integrated classification model that can yield a classifier specifically for industrial applications. Our knowledge incorporated learning scheme enables users to create a classifier that accurately identifies "important samples" whose misclassification can result in severe consequences, while maintaining high precision for the remaining samples. We verify the effectiveness of our proposed method on two real blast furnace datasets, which demonstrates that operators can improve the control of blast furnace systems by leveraging prior knowledge or their experience.